Mortality after deferral of treatment or no treatment for choroidal melanoma

Purpose: To report mortality of patients who were eligible for enrollment in the Collaborative Ocular Melanoma Study (COMS) clinical trials of medium-sized choroidal melanoma or large-sized choroidal melanoma but chose to defer treatment or receive no melanoma treatment. Design: Prospective nonrandomized multicenter cohort study as an adjunct to COMS randomized clinical trials. Methods: Patient follow-up procedures included examinations, correspondence, telephone contacts, and National Death Index searches. Primary outcome was patient death measured by all-cause mortality. Secondary outcomes were melanoma treatment and melanoma metastasis. Results: Of 77 patients eligible for COMS clinical trials who chose to defer or receive no melanoma treatment, 61 were appropriate candidates and 45 (74%) enrolled in the natural history study (NHS). In all, 42 patients (42 eyes) had medium melanoma, and the median follow-up was 5.3 years (range, 4–10.7 years). In all, 22 patients (52%) had subsequent melanoma treatment, and 20 (48%) had no melanoma treatment. For the 42 patients, Kaplan–Meier estimate of 5-year mortality was approximately 30% [95% confidence interval (CI), 18%–47%]. For COMS medium melanoma trial, 5-year mortality was 18% (95% CI, 16%–20%), not statistically significantly different from the NHS patients. After adjusting for differences in age and longest basal diameter, the 5-year risk of death for NHS patients versus COMS trial patients was 1.54 (95% CI, 0.93–2.56). Three patients had large melanoma. Melanoma metastasis was confirmed or suspected in 8 (42%) of 19 deaths. Conclusion: Greater mortality and higher risk of death for NHS patients are probative but not conclusive evidence of a beneficial, life-extending effect of medium melanoma treatment.

Multicollinearity may lead to artificial interaction: an example from a cross sectional study of biomarkers.

Collinearity is the situation which arises in multiple regression when some or all of the explanatory variables are so highly correlated with one another that it becomes very difficult, if not impossible, to disentangle their influences and obtain a reasonably precise estimate of their effects. Suppressor variable is one of the extreme situations of collinearity that one variable can substantially increase the multiple correlation when combined with a variable that is only modestly correlated with the response variable. In this study, we describe the process by which we disentangled and discovered multicollinearity and its consequences, namely artificial interaction, using the data from cross-sectional quantification of several biomarkers. We showed how the collinearity between one biomarker (blood lead level) and another (urinary trans, trans-muconic acid) and their interaction (blood lead level* urinary trans, trans-muconic acid) can lead to the observed artificial interaction on the third biomarker (urinary 5-aminolevulinic acid).